First Proof submissions
An internal OpenAI model attempted all 10 First Proof research-level problems; OpenAI judged at least five likely correct and retracted one belief.
First Proof is a challenge by leading mathematicians requiring end-to-end checkable proofs in specialist domains. OpenAI submitted attempts with limited human supervision, said problems 4, 5, 6, 9 and 10 likely hold, and conceded problem 2 was wrong after official commentary.
- Date
- Friday, 20 February 2026
- Lab
- OpenAI
- Kind
- paper
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| Proof attempts judged likely correct | at least 5 of 10 problems 4, 5, 6, 9, 10; problem 2 later judged incorrect | company |
Self-assessed by OpenAI with expert feedback; official First Proof commentary was not read. Outputs are not a released model.
Sources
This record was checked against its sources on 6 October 2026. How we check